On Fri, 8 Apr 2005, Peter Stuge wrote:
>> I don't know where you get those enormous times from...
>I agree that those times seem long, what does that server run on,
>Andreas? And did you make a special test case application or time
>binc, or some other application? (du?)

It's reiserfs. But not all (even Maildir) mail systems use reiserfs. The 
test I did was on a sleeping system. I didn't run "du" many times first to 
get all entries into the cache.

>> # time du -h --summarize .Mailing\ Lists.Software.Qmail/
>> 43M     .Mailing Lists.Software.Qmail/
>> real    0m0.047s
>> user    0m0.012s
>> sys     0m0.035s
>Kernel caching has huge impact on these test, you would have to make
>sure that nothing in that directory was in the cache before getting
>good numbers. Which can be hard.

Yes, these are definitely cached numbers. Running "du" the second time 
brings my INBOX numbers down from 8.580 secs to 0.015 secs. On a loaded 
mail server, all cache gets purged so you can't benefit from it. ;)

Imagine "du" being run every time someone switches mailboxes on a system 
with more than 100 users... Calculating quota this way renders a mail 
server unusable. :-)

Andy :-)

--
Andreas Aardal Hanssen   | http://www.andreas.hanssen.name/gpg
Author of Binc IMAP      |  "It is better not to do something
http://www.bincimap.org/ |        than to do it poorly."

Reply via email to